Haystack docs home page

Module base

BaseQueryClassifier

class BaseQueryClassifier(BaseComponent)

Abstract class for Query Classifiers

Module sklearn

SklearnQueryClassifier

class SklearnQueryClassifier(BaseQueryClassifier)

A node to classify an incoming query into one of two categories using a lightweight sklearn model. Depending on the result, the query flows to a different branch in your pipeline and the further processing can be customized. You can define this by connecting the further pipeline to either output_1 or output_2 from this node.

Example:

|{
|pipe = Pipeline()
|pipe.add_node(component=SklearnQueryClassifier(), name="QueryClassifier", inputs=["Query"])
|pipe.add_node(component=elastic_retriever, name="ElasticRetriever", inputs=["QueryClassifier.output_2"])
|pipe.add_node(component=dpr_retriever, name="DPRRetriever", inputs=["QueryClassifier.output_1"])

|# Keyword queries will use the ElasticRetriever
|pipe.run("kubernetes aws")

|# Semantic queries (questions, statements, sentences ...) will leverage the DPR retriever
|pipe.run("How to manage kubernetes on aws")

Models:

Pass your own Sklearn binary classification model or use one of the following pretrained ones:

  1. Keywords vs. Questions/Statements (Default) query_classifier can be found here query_vectorizer can be found here output_1 => question/statement output_2 => keyword query Readme

  2. Questions vs. Statements query_classifier can be found here query_vectorizer can be found here output_1 => question output_2 => statement Readme

See also the tutorial on pipelines.

SklearnQueryClassifier.__init__

def __init__(
        model_name_or_path:
    Union[
        str,
        Any] = "https://ext-models-haystack.s3.eu-central-1.amazonaws.com/gradboost_query_classifier/model.pickle",
        vectorizer_name_or_path:
    Union[
        str,
        Any] = "https://ext-models-haystack.s3.eu-central-1.amazonaws.com/gradboost_query_classifier/vectorizer.pickle",
        batch_size: Optional[int] = None,
        progress_bar: bool = True)

Arguments:

  • model_name_or_path: Gradient boosting based binary classifier to classify between keyword vs statement/question queries or statement vs question queries.
  • vectorizer_name_or_path: A ngram based Tfidf vectorizer for extracting features from query.
  • batch_size: Number of queries to process at a time.
  • progress_bar: Whether to show a progress bar.

Module transformers

TransformersQueryClassifier

class TransformersQueryClassifier(BaseQueryClassifier)

A node to classify an incoming query into categories using a transformer model. Depending on the result, the query flows to a different branch in your pipeline and the further processing can be customized. You can define this by connecting the further pipeline to output_1, output_2, ..., output_n from this node. This node also supports zero-shot-classification.

Example:

|{
|pipe = Pipeline()
|pipe.add_node(component=TransformersQueryClassifier(), name="QueryClassifier", inputs=["Query"])
|pipe.add_node(component=elastic_retriever, name="ElasticRetriever", inputs=["QueryClassifier.output_2"])
|pipe.add_node(component=dpr_retriever, name="DPRRetriever", inputs=["QueryClassifier.output_1"])

|# Keyword queries will use the ElasticRetriever
|pipe.run("kubernetes aws")

|# Semantic queries (questions, statements, sentences ...) will leverage the DPR retriever
|pipe.run("How to manage kubernetes on aws")

Models:

Pass your own Transformer classification/zero-shot-classification model from file/huggingface or use one of the following pretrained ones hosted on Huggingface:

  1. Keywords vs. Questions/Statements (Default) model_name_or_path="shahrukhx01/bert-mini-finetune-question-detection" output_1 => question/statement output_2 => keyword query Readme

  2. Questions vs. Statements model_name_or_path="shahrukhx01/question-vs-statement-classifier" output_1 => question output_2 => statement Readme

See also the tutorial on pipelines.

TransformersQueryClassifier.__init__

def __init__(model_name_or_path: Union[
    Path, str] = "shahrukhx01/bert-mini-finetune-question-detection",
             model_version: Optional[str] = None,
             tokenizer: Optional[str] = None,
             use_gpu: bool = True,
             task: str = "text-classification",
             labels: List[str] = DEFAULT_LABELS,
             batch_size: int = 16,
             progress_bar: bool = True,
             use_auth_token: Optional[Union[str, bool]] = None,
             devices: Optional[List[Union[str, torch.device]]] = None)

Arguments:

  • model_name_or_path: Directory of a saved model or the name of a public model, for example 'shahrukhx01/bert-mini-finetune-question-detection'. See Hugging Face models for a full list of available models.
  • model_version: The version of the model to use from the Hugging Face model hub. This can be a tag name, a branch name, or a commit hash.
  • tokenizer: The name of the tokenizer (usually the same as model).
  • use_gpu: Whether to use GPU (if available).
  • task: Specifies the type of classification. Possible values: 'text-classification' or 'zero-shot-classification'.
  • labels: If the task is 'text-classification' and an ordered list of labels is provided, the first label corresponds to output_1, the second label to output_2, and so on. The labels must match the model labels; only the order can differ. If the task is 'zero-shot-classification', these are the candidate labels.
  • batch_size: The number of queries to be processed at a time.
  • progress_bar: Whether to show a progress bar.
  • use_auth_token: The API token used to download private models from Huggingface. If this parameter is set to True, then the token generated when running transformers-cli login (stored in ~/.huggingface) will be used. Additional information can be found here https://huggingface.co/transformers/main_classes/model.html#transformers.PreTrainedModel.from_pretrained
  • devices: List of torch devices (e.g. cuda, cpu, mps) to limit inference to specific devices. A list containing torch device objects and/or strings is supported (For example [torch.device('cuda:0'), "mps", "cuda:1"]). When specifying use_gpu=False the devices parameter is not used and a single cpu device is used for inference.